jabir-khan/hunter-llm-sft-v2

hunter-llm-sft-v2 Curated SFT + DPO dataset for HunterLLM, an authorized-bug-bounty-focused fine-tune of Llama-3-8B-Instruct. What changed vs hunter-llm-sft-v1 v1 paired each prompt with a generic 7-line template as the gold output and stuffed the actual source text into the input. The model learned to emit those templates and ignore the source — shallow answers, factual drift, repetition loops. v2 reverses this: every output is the real source body — the actual write-up… See the full description on the dataset page: huggingface.co/datasets/jabir-khan/hunter-llm-sft-v2.